Social determinants of diabetes-related foot disease among older adults in New South Wales, Australia: evidence from a population-based study

This study has three ethics approvals: NSW Population and Health Services Research Ethics Committee (HREC/13/CIPHS/8), Western Sydney University Human Research Ethics Committee (H12215), and ACT Health Human Research Ethics and Governance (ETHLR.12.173). The 45 and Up Study was approved by the University of New South Wales Human Research Ethics Committee.

This study used data from participants of the 45 and Up Study baseline survey that was conducted between 2006 and 2009 in NSW. The 45 and Up Study was managed by the Sax Institute. About 267,153 respondents were randomly chosen from the Medicare Australia database, national health care records of citizens and permanent residents along with some temporary residents and refugees, who were aged 45 years and older, and had consented to have their medical data linked. This has been detailed elsewhere [24]. The 45 and Up Study is considered the largest population-based cohort study in Australia and the Southern hemisphere. The 45 and Up Study survey responses include self-reported data on participants’ demographic, socioeconomic, lifestyle and health factors.

For our study, the 45 and Up Study survey data were linked with health administrative databases (detailed below) by the Centre for Health Record Linkage (CHeReL) and the Sax Institute, two external agencies that provided the data for each participant deidentified through the use of a person project number (PPN) [25]. The data was accessed through the Secure Unified Research Environment (SURE), a cloud-based platform with a two-step authentication process.

The health administrative datasets linked to the survey data were NSW Admitted Patient Data Collection (APDC), NSW Emergency Department Data Collection (EDDC), and Pharmaceutical Benefits Scheme (PBS). NSW APDC data consists of hospital utilisation data on admitted patients to public hospitals, public psychiatric hospitals, public multi-purpose services, private hospitals, and private day procedures centres between 1999 to 2012 in NSW. EDDC consists of health service data on individuals presented at emergency departments of public hospitals in NSW between 2005 to 2012. In our study, APDC and EDDC of 2006–2012 were used to identify DFD, DFU and DFI. The PBS contains individuals’ data on prescribed pharmaceutical medications from 2004 to 2011. The PBS data were used to classify types of diabetes among survey participants. The CHeReL linked the 45 and Up Study data with APDC and EDDC using a probabilistic matching procedure, whereas the Sax Institute facilitated the linkage of the 45 and Up Study and PBS data using a deterministic matching procedure.

The analytic sample of our study with time period is shown in Fig. 1. Out of 267,153 survey participants, 41 observations were dropped by the Sax Institute due to ongoing maintenance and data cleaning. Our study was supplied with the data of 267,112 participants. A further 26 observations were dropped whose age was less than 45 years old or had missing information. Among 267,086 participants, our study identified 28,210 people with diabetes, and it was the size of our analytic sample. Using this sample, our study utilised APDC and EDDC from 2006 to 2012 to identify people DFD, DFU and DFI.

The term “diabetes-related foot complication” or “diabetic foot complication” was used to refer to any DFD, DFU and DFI conditions in this paper. The identification process of people with diabetes and diabetic foot complications is detailed in Fig. 2. In the first step, diabetes status was determined based on either self-reported data or hospital diagnostic codes [22]. The participants of the 45 and Up Study survey participants were asked, “Has a doctor ever told you that you have diabetes?”. If the answer was “yes”, then they were assigned to have diabetes. In addition, participants belonged to the diabetes group if they were diagnosed with diabetes in the hospital during their hospital admissions or emergency department presentations. Diagnosis codes used in this study were International Classification of Diseases 9 - Clinical Modification (ICD-9 CM), International Classification of Diseases 10 - Australian Modification (ICD-10 AM) and Systematized Nomenclature of Medicine - Clinical Terms (SNOMED- CT). This was due to the different sources of data and the timeframe. The complete list of related diagnostic codes used for this study is presented in Supplementary Table 1 and Supplementary Table 2 [18, 26‐30].

In the second step, the identification of diabetes-related foot complications was ascertained only if the foot complication was identified at the same time or after diabetes identification (Fig. 2).

The analytic sample consists of 28,210 individuals whose diabetic foot complications status were observed in APDC and EDDC from 2006 to 2012. The study outcome variables were DFD, DFU and DFI, which were dichotomised. DFD included the ulcer of foot or lower limb, decubitus ulcer, peripheral angiopathy with or without gangrene, cellulitis of toe or lower limb, osteomyelitis, mono/polyneuropathy of lower limb, neuropathic arthropathy, and diabetes-related amputation of the lower limb [1, 18, 19, 31]. DFU included ulcer of foot or lower limb, decubitus ulcer, whereas DFI included cellulitis or osteomyelitis of foot or lower limb [1, 32, 33].

Our choice of study variables was based on similar studies that examined the relationship between DFD (or diabetes) and other factors [19‐22, 34‐40]. For instance, the study by Lazzarini et al. (2017) included age, sex, socioeconomic status, education level, smoking status and depression while investigating the associated factors of foot complications in a representative inpatient population of Australia [20]. They found that age and socioeconomic status were significantly associated with a previous foot ulcer. Tapp et al. (2002) concluded that diabetes duration and hypertension were predictors of peripheral vascular disease in an Australian population-based study [22]. The study by Nather et al. (2010) demonstrated that household income and physical activity to be significantly associated with diabetic foot in an inpatient study in Singapore [37]. Study variables of our study were classified into demographic, socioeconomic, lifestyle risk factors, and health status variables.

Demographic variables included sex, age, marital status, remoteness of residence, country of birth, and whether they spoke a language other than English. Socioeconomic variables included the highest education level, socioeconomic status, household income, and private health insurance status. These variables were categorised following previous Australian studies [20, 37, 41‐43]. To mention a few, age was categorised into four groups: 45–54 years, 55–64 years, 65–74 years, and 75 years or older following the 45 and Up Study related diabetes studies [41, 42, 44]. The highest attained educational qualification had four groups: less than high school, high school certificate/trade, certificate/diploma, and university and higher [43]. Socioeconomic status was assessed by the Australian Bureau of Statistics’ index of relative socio-economic disadvantage (IRSD) [45]. In our study, IRSD scores (the highest proportion of disadvantage) was ranked in quantiles, with lower scores indicating higher levels of social disadvantage [43]. Annual household income (in AUD) was classified into three categories: less than AUD 20,000, AUD 20,000 to less than AUD 50,000, and AUD 50,000 and over [41, 44]. The private health insurance variable was categorised into three groups: “No (without Department of Veterans’ Affairs (DVA) card and concession card)”, “No (with DVA card or concession card) and “Yes”. The current or former members of the Australian Defence Force or their dependents are eligible for DVA card, whereas eligible people can access the concession card. The eligibility of concession cards is determined by many factors, including age, income, and disability status. DVA or concession cardholders can receive more subsidised health services and medicine compared to non-holders [46, 47].

Lifestyle risk factor variables included self-reported smoking status, alcohol consumption, physical activity, and fruits and vegetables intake. The classification followed previous studies or guidelines [48‐50]. Health status variables included the duration of diabetes, body mass index (BMI), presence of different comorbidities (high blood pressure, high blood cholesterol, heart disease, stroke, asthma, and psychological distress). BMI was calculated from self-reported height and weight and was categorised according to the National Health and Medical Research Council guidelines [51]. Psychological distress, measured using the Kessler-10 (K10) instrument [52] were classified into two groups: “None/low/moderate” and “High/very high” [43].

Survey logistic regression adjusted for survey weights to determine the association between each outcome measure and different risk factors. In the first step, univariate analysis was performed to examine the unadjusted odds ratio (OR). Next, multivariate logistic regression models were employed to examine the association between each outcome variable and study factors. Adjusted Odds ratios (AORs) with 95% confidence interval (CI) were calculated to measure the association of the study variables and DFD and its different types.

The potential impact of missing data was considered on the estimated coefficients in sensitivity analyses with the imputed dataset. The chain equation method was applied for the multiple imputations, assuming that data were missing at random [53]. The approach also assumed that available information on the participants’ characteristics could be used to investigate the participants with missing data [54]. All study factors and outcome variables of the main analysis were considered in the multiple imputation models. The imputation was conducted using Stata 16 with ‘mi’ command. The sensitivity analysis was conducted based on 25 imputations [55], and revised AORs with 95% CI were presented to compare with complete case analysis.

The present study included 28,210 people with diabetes aged 45 years and older in NSW, Australia from the 45 and Up Study data. The number of people identified with DFD was 3035 during 2006–2012. Among 3035 individuals with DFD, 838 individuals were diagnosed with DFU only, 783 individuals were diagnosed with DFI only, 665 persons had both DFU and DFI, while 741 persons had other types of DFD (peripheral vascular disease, peripheral neuropathy, mononeuropathy, and diabetic neuropathic arthropathy).

The number of people with DFD, DFU and DFI was also estimated at the population level (weighted frequency). The total number of DFD patients were estimated as 33,663 in NSW during 2006–2012 at the population level, whereas the estimated number of DFU and DFI patients were 16,976 and 16,248, respectively (Fig. 3). As shown in Fig. 3, there was an overlap between DFU and DFI (2.4%) adults with diabetes aged 45 years and older who lived in NSW. Again, 10.8% of adults with diabetes aged 45 years and older from NSW had DFD, 5.4% had DFU, and 5.2% had DFI (Fig. 3).

Participants’ demographic, socioeconomic, health status, and lifestyle characteristics (weighted frequency and weighted percentage) is presented in Table 1. A large number of participants (40–45%) with DFD, DFU and DFI belonged to the “75 years or older age group” while most were from English-speaking countries (75–80%). About two-thirds of the participants were from households with an annual income of less than AUD 20,000 for all types of foot complications. Regarding lifestyle factors, about 55% of survey participants were reported to be ever being a regular smoker, and 54–58% of participants performed less than 150 min of moderate-to-vigorous physical activity per week. In regard to health status factors, it was found that 39–50% of diabetic foot patients had diabetes for 15 years or more, and around 65% of patients suffered cardiovascular diseases.

The association of study factors and DFD in terms of unadjusted and adjusted ORs from logistic regression with 95% CI was presented in Table 2. Regarding demographic factors, the study showed that older people (65–74 years and 75 years or over), males, single individuals, people from English-speaking countries were significantly more likely to have DFD. Individuals from remote areas had an 82% higher likelihood of experiencing DFD than those in major cities (p = 0.007). Individuals with an annual household income of AUD 20,000 to less than AUD 50,000 had significantly lower likelihood (AOR = 0.83, 95% CI: 0.70, 0.98) of having DFD compared to individuals with less than AUD 20,000 household income. The odds of having DFD was also found to be lower for people with income of AUD 50,000 or more (AOR = 0.59, 95% CI: 0.46, 0.76).

Smoking and physical activity were significantly associated with having DFD (p < 0.001). It was observed that the odds of experiencing DFD was higher among people who were ever being a regular smoker (AOR = 1.31, 95% CI: 1.13, 1.51) compared to those who were never a regular smoker (Table 2). Again, people who performed moderate-to-vigorous physical activity of 150 min or more per week had a lower likelihood of having DFD (AOR = 0.63, 95% CI: 0.51, 0.78 and AOR = 0.59, 95% CI: 0.50, 0.68). Health status factors associated with a higher likelihood of having DFD included type-1 diabetes, diabetes duration of 15 years or more, and lower BMI. Results also indicated that people were more likely to have DFD if they had cardiovascular disease (AOR = 1.62, 95% CI: 1.40, 1.89), had a stroke (AOR = 1.37, 95% CI: 1.09, 1.73), had high level of psychological distress (AOR = 1.30, 95% CI: 1.03, 1.65) as opposed to those who did not have these conditions.

The unadjusted and adjusted ORs for the association between DFU and study variables are shown in Table 3. With regards to demographic and socioeconomic factors, older age, male, single, individuals from English speaking countries, and patients from lower household income had a significantly higher likelihood of having DFU. Results showed that people aged 75 years and over (AOR = 4.19, 95% CI: 2.66, 6.58) were more likely to experience DFD than people from the 45–54 years age group. People who reported having an annual household income of more than AUD 50,000 (AOR = 0.58, 95% CI: 0.40, 0.84) had a lower likelihood of having DFU compared to people with a household income of less than AUD 20,000.

The unadjusted and adjusted ORs for the association between DFI and study factors were illustrated in Table 4. Demographic and socioeconomic factors associated with having DFI were older age, male, single status, coming from lower household income, and having no private health insurance. People aged 75 years or more (AOR = 1.93, 95% CI: 1.28, 2.90) were more likely to experience DFI than people of age group 45–54 years. The study found that people living in the most disadvantaged areas (AOR = 1.42, 95% CI: 1.02, 1.97) had a higher likelihood of suffering from DFI than people from the least disadvantaged group. It was also found that people with household income of AUD 50,000 (AOR = 0.62, 95% CI: 0.44, 0.87) had lower odds of having DFI than people with a household income of less than AUD 20,000. Again, people who had private health insurance (AOR = 0.71, 95% CI: 0.52, 0.94) were significantly less likely to have DFI compared to those who did not have private health insurance and DVA/concession card.

Although the complete case analysis did not find any association of remoteness of residence area, country of birth and education with DFI (Table 4), the multiple imputation analysis suggested that the associations were significant (Supplementary Table 3). Results from multiple imputation analyses suggested that people residing in remote areas (AOR = 1.65, 95% CI: 1.05, 2.58) were more likely to have DFI than major city dwellers (Supplementary Table 3).

Among the lifestyle factors considered in this study, physical activity was found significantly associated with DFI. People who performed moderate-to-vigorous physical activity for 150–300 min per week or more were significantly less likely (AOR = 0.70, 95% CI: 0.53, 0.93; and AOR = 0.64, 95% CI: 0.52, 0.80) to experience DFI compared to people who performed similar physical activity for less than 150 min per week. Regarding health status variables, the study found a significant association of DFI with type-1 diabetes, longer diabetes duration, stroke, and asthma. Compared to people with type-1 diabetes, people with type-2 diabetes (AOR = 0.55, 95% CI: 0.33, 0.90) had lower odds of suffering from DFI. The likelihood of having DFI was significantly higher among people with diabetes duration 15 years or more (AOR = 1.62, 95% CI: 1.25, 2.10) compared to people with less than 5 years of diabetes duration. Results also indicated that people who had cardiovascular disease, had a stroke, or had asthma were more likely (AOR = 1.36, 95% CI: 1.11, 1.67, AOR = 1.4, 95% CI: 1.01, 1.94 and, AOR = 1.64, 95% CI: 1.26, 2.15, respectively) to have DFI compared to those who did not have the condition.

The revised AORs did not differ markedly from the complete case analysis (Supplementary Table 3), which indicated that missing data did not substantially affect the findings from the observed data.

This study identified individuals with diabetes based on responses to self-reported surveys, emergency department presentation diagnosis codes and/or hospital admission. One of the strengths of this study is that the identification of diabetes is quite robust in this study compared to those studies which relied only on self-reported diabetes status or only used hospital admission diagnosis codes. Another strength of the study is its’ provision of insights into social factors related to DFD for the older population of NSW using a representative sample that was not previously explored.

The study has a few limitations as well. Firstly, the study could not ascertain the causal relationship between the study factors and the outcome variables due to the cross-sectional data collection. Secondly, the study could not rule out the possibility of underestimation of diabetes, and consequently DFDs in the absence of clinical data on diabetes during the survey. However, fact remains, clinical data is not always available for a large-scale study like the 45 and Up Study. Since the survey data were linked with administrative health data like APDC, EDDC and PBS, it is reasonable to infer that the information contained in the study could be of good use to ascertain a person’s diabetes status much better than those studies that were only based on self-reported data. Thirdly, the coding in ICD-10 AM, ICD-9 CM and SNOMED are prone to human error [65]. If the foot diseases were coded wrong or were not coded at all during admission, the estimate would result in underestimating DFD and its different types. However, one study reported that the accuracy of coding was acceptable to make reliable estimates [18]. Fourthly, the period to which our study related is not the most recent, but this does not diminish the strength of the findings. To our knowledge, this data is current for NSW because time-periods are associated with large data linkage-based studies due to the time involved in obtaining, training, linking, cleaning and validating cross-system data linkage. Lastly, there is a possibility of including leg ulcers in this study that were not precisely limited to the foot due to hospital coding. Despite these limitations, a major strength of the study is using a large representative survey linked with administrative health data that identifies individuals with diabetes and DFDs.